China has nearly banned the use of Western accelerators for AI across various sectors, yet, according to the Financial Times, Chinese giants are in no hurry to abandon Nvidia’s developments. 
Created by Gemini. The report indicates that Chinese tech giants, including Alibaba and ByteDance, are increasingly training their most advanced artificial intelligence models using Nvidia accelerators. However, these are not purchased for their needs but accessed remotely via cloud data centers in Southeast Asia. This showcases how leading Chinese companies are circumventing US export controls and simultaneously bypassing China’s own bans.
Over the past year, Alibaba’s large language models, Qwen, and ByteDance’s Doubao have risen to the top echelons of global LLM benchmarks. It is believed that both models have been trained, at least in part, using Nvidia accelerators located in overseas cloud data centers. Simultaneously, operators of these data centers from Singapore say demand from Chinese companies has increased since April. That was precisely when the Donald Trump administration imposed tighter embargoes on Nvidia’s H20 chips.
Amid these regulatory challenges, Nvidia has managed to maintain its appeal due to its superior AI processing capabilities, which remain unmatched by local Chinese alternatives. The recent strategic pivots in US policy continue to escalate global semiconductor tensions. Meanwhile, China’s long-term strategy focuses on advancing its domestic chip production to lessen this dependency, highlighting the ongoing tech race between the two superpowers.
